计算机工程与应用 ›› 2013, Vol. 49 ›› Issue (19): 167-172.

• 图形图像处理 • 上一篇    下一篇

基于轮廓与SIFT特征组合的商标图像检索

向  雷1,肖诗斌1,2,林春雨1,2,吕学强1,2   

  1. 1.北京信息科技大学 中文信息处理研究中心,北京 100101
    2.北京拓尔思信息技术股份有限公司,北京 100101
  • 出版日期:2013-10-01 发布日期:2015-04-20

Trademark images retrieval based on SIFT feature and profile feature

XIANG Lei1, XIAO Shibin1,2, LIN Chunyu1,2, LV Xueqiang1,2   

  1. 1.Chinese Information Processing Research Center, Beijing Information Science and Technology University, Beijing 100101, China
    2.Beijing TRS Information Technology Co., Ltd., Beijing 100101, China
  • Online:2013-10-01 Published:2015-04-20

摘要: 针对单一特征对商标图像描述的局限性,提出了一种基于轮廓和SIFT特征组合的商标图像检索方法。该方法对二值化的商标图像进行轮廓提取,采用规则算法对其进行轮廓分解,对分解的参考点集进行Fourier变换,将得到的Fourier系数作为参考点的轮廓特征。针对商标图像的尺度空间进行极值点检测,并对检测到的极值点进行特征描述,该特征描述即为商标图像的SIFT特征描述。最后,SIFT特征与轮廓特征进行特征融合,并将融合后的组合特征作为对商标图像的特征描述。

关键词: 轮廓特征, SIFT特征, 轮廓提取, 轮廓分解, Fourier变换

Abstract: For the limitations of the description on single feature trademark image, a method of trademark image retrieval which is based on profile and SIFT characteristics is proposed. In this method, first step is to get profile by making trademark image to binary image, then contour decomposes the image using rule algorithm and does the Fourier transform on set of reference points of the decomposition. The Fourier coefficient is the profile feature of reference point. Then it detects extreme points on scale space of the trademark image, and describes the extreme points with feature which is SIFT characteristics description of trademark image. Finally, SIFT features is combined with profile features as a new feature for feature description of trademark image.

Key words: profile feature, SIFT feature, contour extraction, profile decomposition, Fourier transform